Re: [openstack-dev] stalled bug fixes for vmware driver
Thanks to everyone for the reviews over the last 24 hours. I really appreciate the time and effort from all in the review process. I hope that once we have the tempest results posted automatically for each for each patch then it will give the core reviewers more confidence in the Vmware fixes. Thanks and have a good weekend. Gary On 9/21/13 3:10 AM, Russell Bryant rbry...@redhat.com wrote: On 09/20/2013 07:00 PM, Dan Wendlandt wrote: btw, thanks to the core devs who went and took a look at several of the vmware reviews today. It was like christmas for the team today :) And for the record, I did all of my reviews before this thread even started, just as a part of my normal workflow. I was working from the havana-rc1 bug list. :-) -- Russell Bryant ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
Re: [openstack-dev] stalled bug fixes for vmware driver
On 09/20/2013 04:11 PM, Dan Wendlandt wrote: Hi Russell, Thanks for the detailed thoughts. Comments below, Dan On Fri, Sep 20, 2013 at 11:52 AM, Russell Bryant rbry...@redhat.com mailto:rbry...@redhat.com wrote: On 09/20/2013 02:02 PM, Dan Wendlandt wrote: I think the real problem here is that in Nova there are bug fixes that are tiny and very important to a particular subset of the user population and yet have been around for well over a month without getting a single core review. Take for example https://review.openstack.org/#/c/40298/ , which fixes an important snapshot bug for the vmwareapi driver. This was posted well over a month ago on August 5th. It is a solid patch, is 54 new/changed lines including unit test enhancements. The commit message clearly shows which tempest tests it fixes. It has been reviewed by many vmware reviewers with +1s for a long time, but the patch just keeps having to be rebased as it sits waiting for core reviewer attention. To me, the high-level take away is that it is hard to get new contributors excited about working on Nova when their well-written and well-targeted bug fixes just sit there, getting no feedback and not moving closer to merging. The bug above was the developer's first patch to OpenStack and while he hasn't complained a bit, I think the experience is far from the community behavior that we need to encourages new, high-quality contributors from diverse sources. For Nova to succeed in its goals of being a platform agnostic cloud layer, I think this is something we need a community strategy to address and I'd love to see it as part of the discussion put forward by those people nominating themselves as PTL. I've discussed this topic quite a bit in the past. In short, my approach has been: 1) develop metrics 2) set goals 3) track progress against those goals The numbers I've been using are here: http://russellbryant.net/openstack-stats/nova-openreviews.html Its great that you have dashboards like this, very cool. The interesting thing here is that the patches I am talking about are not waiting on reviews in general, but rather core review. They have plenty of reviews from non-core folks who provide feedback (and they keep getting +1'd again as they are rebased every few days). Perhaps a good additional metric to track would be be items that have spent a lot of time without a negative review, but have not gotten any core reviews. I think that is the root of the issue in the case of the reviews I'm talking about. The numbers I track do not reset the timer on any +1 (or +2, actually). I only resets when it gets a -1 or -2. At that point, the review is waiting for an update from a submitter. Point is, getting a bunch of +1s does not make it show up lower on the list. Also, the 3rd list (time since the last -1) does not reset on a rebase, so that's covered in this tracking, too. -- Russell Bryant ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
Re: [openstack-dev] stalled bug fixes for vmware driver
Hi Russell, Thanks for the detailed thoughts. Comments below, Dan On Fri, Sep 20, 2013 at 11:52 AM, Russell Bryant rbry...@redhat.com wrote: On 09/20/2013 02:02 PM, Dan Wendlandt wrote: I think the real problem here is that in Nova there are bug fixes that are tiny and very important to a particular subset of the user population and yet have been around for well over a month without getting a single core review. Take for example https://review.openstack.org/#/c/40298/ , which fixes an important snapshot bug for the vmwareapi driver. This was posted well over a month ago on August 5th. It is a solid patch, is 54 new/changed lines including unit test enhancements. The commit message clearly shows which tempest tests it fixes. It has been reviewed by many vmware reviewers with +1s for a long time, but the patch just keeps having to be rebased as it sits waiting for core reviewer attention. To me, the high-level take away is that it is hard to get new contributors excited about working on Nova when their well-written and well-targeted bug fixes just sit there, getting no feedback and not moving closer to merging. The bug above was the developer's first patch to OpenStack and while he hasn't complained a bit, I think the experience is far from the community behavior that we need to encourages new, high-quality contributors from diverse sources. For Nova to succeed in its goals of being a platform agnostic cloud layer, I think this is something we need a community strategy to address and I'd love to see it as part of the discussion put forward by those people nominating themselves as PTL. I've discussed this topic quite a bit in the past. In short, my approach has been: 1) develop metrics 2) set goals 3) track progress against those goals The numbers I've been using are here: http://russellbryant.net/openstack-stats/nova-openreviews.html Its great that you have dashboards like this, very cool. The interesting thing here is that the patches I am talking about are not waiting on reviews in general, but rather core review. They have plenty of reviews from non-core folks who provide feedback (and they keep getting +1'd again as they are rebased every few days). Perhaps a good additional metric to track would be be items that have spent a lot of time without a negative review, but have not gotten any core reviews. I think that is the root of the issue in the case of the reviews I'm talking about. There is also an aspect of karma involved in all of this that I think phttp://summit.openstack.org/cfp/details/4lays a big part when it comes to the vmware driver patches. To get review attention, someone has to *want* to review it. To get people to want to review your stuff, it can either be technology they are interested in, or they just want to help you out personally. There aren't many Nova developers that use the vmware driver, so you've got to work on building karma with the team. That means contributing to other areas, which honestly, I haven't seen much of from this group. I think that would go a long way. I agree with you on the dynamics of review karma here (having dealt with this same issue as a PTL of Quantum/Neutron). My sense of of is going on here is a bootstrapping issue. If you have a developer who is brand new to OpenStack, it would make sense that your first patch or two would be in an area where you feel most comfortable (e.g., because you understand the VMware APIs + constructs). For new developers, who aren't pushing a big new feature but are instead just fixing an existing bug, I would not guessed that a huge amount of karma is needed for a review. People like garyk and arosen who have more Nova experience are already doing Nova work outside of the VMware driver (instance groups, neutron / security groups code, many reviews throughout Nova, not to mention their work in Neutron) and I expect that to be the path others follow as well. Nonetheless, I like the suggestion of having these new developers also try to gain experience outside of the vmware driver and provide value to the wider community... is https://bugs.launchpad.net/nova/+bugs?field.tag=low-hanging-fruit still the best place for them to look? Doesn't seem to be much there that isn't already in progress. I think if you review the history of vmware patches, you'll see my name as a reviewer on many (or perhaps most) of them, so I hope nobody thinks that I personally am trying stall things here. This is just based on my experience across all contributions. Indeed, and in fact, when I first wrote the email, I had called out you, Dan Smith, and Michael Still as having been very helpful with VMware API review, but then I was worried that I had left people off the list that had also been helpful but weren't immediately coming to mind. I guess I was damned if I did and damned if I didn't :) My goal in sending the email was not to
Re: [openstack-dev] stalled bug fixes for vmware driver
On Fri, Sep 20, 2013 at 11:52 AM, Russell Bryant rbry...@redhat.com wrote: On 09/20/2013 02:02 PM, Dan Wendlandt wrote: I think the real problem here is that in Nova there are bug fixes that are tiny and very important to a particular subset of the user population and yet have been around for well over a month without getting a single core review. Take for example https://review.openstack.org/#/c/40298/ , which fixes an important snapshot bug for the vmwareapi driver. This was posted well over a month ago on August 5th. It is a solid patch, is 54 new/changed lines including unit test enhancements. The commit message clearly shows which tempest tests it fixes. It has been reviewed by many vmware reviewers with +1s for a long time, but the patch just keeps having to be rebased as it sits waiting for core reviewer attention. Personally I tend not to review many vmwareapi patches because without seeing any public functional tests or being able to run the patch myself, I am uncomfortable saying it 'looks good to me'. All I can do is make sure the code looks pythonic and make no assessment on if the patch works or not. With no shortage of patches to review I tend to review other patches instead. I while back Russell announced we would like all virt drivers have a public functional testing system by the release of Icehouse ( http://lists.openstack.org/pipermail/openstack-dev/2013-July/011280.html). Public functional testing would allow me to review vmwareapi patches with almost the same level of confidence as with a driver that we gate on and that I can trivially try out, such as libvirt. Until then, if we put an explicit comment in the release notes explicitly saying that vmwareapi is a group C virt driver (https://wiki.openstack.org/wiki/HypervisorSupportMatrix -- These drivers have minimal testing and may or may not work at any given time. Use them at your own risk) that would address my concerns and I would be happy to +2 vmwareapi patches based just on if the code looks correct and not on how well the patch works. To me, the high-level take away is that it is hard to get new contributors excited about working on Nova when their well-written and well-targeted bug fixes just sit there, getting no feedback and not moving closer to merging. The bug above was the developer's first patch to OpenStack and while he hasn't complained a bit, I think the experience is far from the community behavior that we need to encourages new, high-quality contributors from diverse sources. For Nova to succeed in its goals of being a platform agnostic cloud layer, I think this is something we need a community strategy to address and I'd love to see it as part of the discussion put forward by those people nominating themselves as PTL. I've discussed this topic quite a bit in the past. In short, my approach has been: 1) develop metrics 2) set goals 3) track progress against those goals The numbers I've been using are here: http://russellbryant.net/openstack-stats/nova-openreviews.html Right now we're running a bit behind the set goal of keeping the average under 5 days for the latest revision (1st set of numbers), and 7 days for the oldest revision since the last -1 (3rd set of numbers). However, Nova is now (and has been since I've been tracking this) below the average for all OpenStack projects: http://russellbryant.net/openstack-stats/all-openreviews.html Review prioritization is not something that I or anyone else can strictly control, but we can provide tools and guidelines to help. You can find some notes on that here: https://wiki.openstack.org/wiki/Nova/CoreTeam#Review_Prioritization There is also an aspect of karma involved in all of this that I think phttp://summit.openstack.org/cfp/details/4lays a big part when it comes to the vmware driver patches. To get review attention, someone has to *want* to review it. To get people to want to review your stuff, it can either be technology they are interested in, or they just want to help you out personally. There aren't many Nova developers that use the vmware driver, so you've got to work on building karma with the team. That means contributing to other areas, which honestly, I haven't seen much of from this group. I think that would go a long way. I think if you review the history of vmware patches, you'll see my name as a reviewer on many (or perhaps most) of them, so I hope nobody thinks that I personally am trying stall things here. This is just based on my experience across all contributions. I already put a session on the design summit schedule to discuss the future of drivers. I'm open to alternative approaches for driver maintenance, including moving some of them (such as the vmware driver) into another tree where the developers focused on it can merge their code without waiting for nova-core review. http://summit.openstack.org/cfp/details/4
Re: [openstack-dev] stalled bug fixes for vmware driver
On 09/20/2013 01:24 PM, Dan Wendlandt wrote: On Fri, Sep 20, 2013 at 1:09 PM, Joe Gordon joe.gord...@gmail..com mailto:joe.gord...@gmail.com wrote: On Fri, Sep 20, 2013 at 11:52 AM, Russell Bryant rbry...@redhat.com mailto:rbry...@redhat.com wrote: On 09/20/2013 02:02 PM, Dan Wendlandt wrote: I think the real problem here is that in Nova there are bug fixes that are tiny and very important to a particular subset of the user population and yet have been around for well over a month without getting a single core review. Take for example https://review.openstack.org/#/c/40298/ , which fixes an important snapshot bug for the vmwareapi driver. This was posted well over a month ago on August 5th. It is a solid patch, is 54 new/changed lines including unit test enhancements. The commit message clearly shows which tempest tests it fixes. It has been reviewed by many vmware reviewers with +1s for a long time, but the patch just keeps having to be rebased as it sits waiting for core reviewer attention. Personally I tend not to review many vmwareapi patches because without seeing any public functional tests or being able to run the patch myself, I am uncomfortable saying it 'looks good to me'. All I can do is make sure the code looks pythonic and make no assessment on if the patch works or not. With no shortage of patches to review I tend to review other patches instead. I while back Russell announced we would like all virt drivers have a public functional testing system by the release of Icehouse (http://lists.openstack.org/pipermail/openstack-dev/2013-July/011280.html). Public functional testing would allow me to review vmwareapi patches with almost the same level of confidence as with a driver that we gate on and that I can trivially try out, such as libvirt. Until then, if we put an explicit comment in the release notes explicitly saying that vmwareapi is a group C virt driver (https://wiki.openstack.org/wiki/HypervisorSupportMatrix -- These drivers have minimal testing and may or may not work at any given time. Use them at your own risk) that would address my concerns and I would be happy to +2 vmwareapi patches based just on if the code looks correct and not on how well the patch works. Hi Joe, I couldn't agree more. In fact, the VMware team has been working hard to get a fully-automated CI infrastructure setup and integrated with upstream Gerrit. We already run the tempest tests internally and have been manually posting tempest results for some patches. I wouldn't want to speak for the dev owner, but I think within a very short time (before Havana) you will begin seeing automated reports for tempest tests on top of vSphere showing up on Gerrit. I agree that this will really help core reviewers gain confidence that not only does the code look OK, but that it works well too. WOOT! To me, the high-level take away is that it is hard to get new contributors excited about working on Nova when their well-written and well-targeted bug fixes just sit there, getting no feedback and not moving closer to merging. The bug above was the developer's first patch to OpenStack and while he hasn't complained a bit, I think the experience is far from the community behavior that we need to encourages new, high-quality contributors from diverse sources. For Nova to succeed in its goals of being a platform agnostic cloud layer, I think this is something we need a community strategy to address and I'd love to see it as part of the discussion put forward by those people nominating themselves as PTL. I've discussed this topic quite a bit in the past. In short, my approach has been: 1) develop metrics 2) set goals 3) track progress against those goals The numbers I've been using are here: http://russellbryant.net/openstack-stats/nova-openreviews.html Right now we're running a bit behind the set goal of keeping the average under 5 days for the latest revision (1st set of numbers), and 7 days for the oldest revision since the last -1 (3rd set of numbers). However, Nova is now (and has been since I've been tracking this) below the average for all OpenStack projects: http://russellbryant.net/openstack-stats/all-openreviews.html http://russellbryant.net/openstack-stats/all-openreviews..html Review prioritization is not something that I or anyone
Re: [openstack-dev] stalled bug fixes for vmware driver
On Fri, Sep 20, 2013 at 1:09 PM, Joe Gordon joe.gord...@gmail.com wrote: On Fri, Sep 20, 2013 at 11:52 AM, Russell Bryant rbry...@redhat.comwrote: On 09/20/2013 02:02 PM, Dan Wendlandt wrote: I think the real problem here is that in Nova there are bug fixes that are tiny and very important to a particular subset of the user population and yet have been around for well over a month without getting a single core review. Take for example https://review.openstack.org/#/c/40298/ , which fixes an important snapshot bug for the vmwareapi driver. This was posted well over a month ago on August 5th. It is a solid patch, is 54 new/changed lines including unit test enhancements. The commit message clearly shows which tempest tests it fixes. It has been reviewed by many vmware reviewers with +1s for a long time, but the patch just keeps having to be rebased as it sits waiting for core reviewer attention. Personally I tend not to review many vmwareapi patches because without seeing any public functional tests or being able to run the patch myself, I am uncomfortable saying it 'looks good to me'. All I can do is make sure the code looks pythonic and make no assessment on if the patch works or not. With no shortage of patches to review I tend to review other patches instead. I while back Russell announced we would like all virt drivers have a public functional testing system by the release of Icehouse ( http://lists.openstack.org/pipermail/openstack-dev/2013-July/011280.html). Public functional testing would allow me to review vmwareapi patches with almost the same level of confidence as with a driver that we gate on and that I can trivially try out, such as libvirt. Until then, if we put an explicit comment in the release notes explicitly saying that vmwareapi is a group C virt driver ( https://wiki.openstack.org/wiki/HypervisorSupportMatrix -- These drivers have minimal testing and may or may not work at any given time. Use them at your own risk) that would address my concerns and I would be happy to +2 vmwareapi patches based just on if the code looks correct and not on how well the patch works. Hi Joe, I couldn't agree more. In fact, the VMware team has been working hard to get a fully-automated CI infrastructure setup and integrated with upstream Gerrit. We already run the tempest tests internally and have been manually posting tempest results for some patches. I wouldn't want to speak for the dev owner, but I think within a very short time (before Havana) you will begin seeing automated reports for tempest tests on top of vSphere showing up on Gerrit. I agree that this will really help core reviewers gain confidence that not only does the code look OK, but that it works well too. Dan To me, the high-level take away is that it is hard to get new contributors excited about working on Nova when their well-written and well-targeted bug fixes just sit there, getting no feedback and not moving closer to merging. The bug above was the developer's first patch to OpenStack and while he hasn't complained a bit, I think the experience is far from the community behavior that we need to encourages new, high-quality contributors from diverse sources. For Nova to succeed in its goals of being a platform agnostic cloud layer, I think this is something we need a community strategy to address and I'd love to see it as part of the discussion put forward by those people nominating themselves as PTL. I've discussed this topic quite a bit in the past. In short, my approach has been: 1) develop metrics 2) set goals 3) track progress against those goals The numbers I've been using are here: http://russellbryant.net/openstack-stats/nova-openreviews.html Right now we're running a bit behind the set goal of keeping the average under 5 days for the latest revision (1st set of numbers), and 7 days for the oldest revision since the last -1 (3rd set of numbers). However, Nova is now (and has been since I've been tracking this) below the average for all OpenStack projects: http://russellbryant.net/openstack-stats/all-openreviews.html Review prioritization is not something that I or anyone else can strictly control, but we can provide tools and guidelines to help. You can find some notes on that here: https://wiki.openstack.org/wiki/Nova/CoreTeam#Review_Prioritization There is also an aspect of karma involved in all of this that I think phttp://summit.openstack.org/cfp/details/4lays a big part when it comes to the vmware driver patches. To get review attention, someone has to *want* to review it. To get people to want to review your stuff, it can either be technology they are interested in, or they just want to help you out personally. There aren't many Nova developers that use the vmware driver, so you've got to work on building karma with the team. That means contributing to other areas, which
Re: [openstack-dev] stalled bug fixes for vmware driver
On Fri, Sep 20, 2013 at 1:37 PM, Russell Bryant rbry...@redhat.com wrote: On 09/20/2013 04:11 PM, Dan Wendlandt wrote: Hi Russell, Thanks for the detailed thoughts. Comments below, Dan On Fri, Sep 20, 2013 at 11:52 AM, Russell Bryant rbry...@redhat.com mailto:rbry...@redhat.com wrote: On 09/20/2013 02:02 PM, Dan Wendlandt wrote: I think the real problem here is that in Nova there are bug fixes that are tiny and very important to a particular subset of the user population and yet have been around for well over a month without getting a single core review. Take for example https://review.openstack.org/#/c/40298/ , which fixes an important snapshot bug for the vmwareapi driver. This was posted well over a month ago on August 5th. It is a solid patch, is 54 new/changed lines including unit test enhancements. The commit message clearly shows which tempest tests it fixes. It has been reviewed by many vmware reviewers with +1s for a long time, but the patch just keeps having to be rebased as it sits waiting for core reviewer attention. To me, the high-level take away is that it is hard to get new contributors excited about working on Nova when their well-written and well-targeted bug fixes just sit there, getting no feedback and not moving closer to merging. The bug above was the developer's first patch to OpenStack and while he hasn't complained a bit, I think the experience is far from the community behavior that we need to encourages new, high-quality contributors from diverse sources. For Nova to succeed in its goals of being a platform agnostic cloud layer, I think this is something we need a community strategy to address and I'd love to see it as part of the discussion put forward by those people nominating themselves as PTL. I've discussed this topic quite a bit in the past. In short, my approach has been: 1) develop metrics 2) set goals 3) track progress against those goals The numbers I've been using are here: http://russellbryant.net/openstack-stats/nova-openreviews.html Its great that you have dashboards like this, very cool. The interesting thing here is that the patches I am talking about are not waiting on reviews in general, but rather core review. They have plenty of reviews from non-core folks who provide feedback (and they keep getting +1'd again as they are rebased every few days). Perhaps a good additional metric to track would be be items that have spent a lot of time without a negative review, but have not gotten any core reviews. I think that is the root of the issue in the case of the reviews I'm talking about. The numbers I track do not reset the timer on any +1 (or +2, actually). I only resets when it gets a -1 or -2. At that point, the review is waiting for an update from a submitter. Point is, getting a bunch of +1s does not make it show up lower on the list. Also, the 3rd list (time since the last -1) does not reset on a rebase, so that's covered in this tracking, too. I see, I misunderstood the labels. One thing to consider adding would be something measuring the patches that have gone the longest without any core review, which (by my current understanding) isn't currently measured. Again, I think its great that you have these charts, and I'm quite sure that its because of your use to charts like this that helped you be so helpful in spotting reviewers that are stalled in the vmware driver and elsewhere. Thanks again for your help on that front. Dan -- Russell Bryant ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev -- ~~~ Dan Wendlandt Nicira, Inc: www.nicira.com twitter: danwendlandt ~~~ ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
Re: [openstack-dev] stalled bug fixes for vmware driver
What criteria would be used to determine which drivers stay in-tree vs. maintained as forks? E.g. libvirt driver in, everyone else out? Open-platform drivers (libvirt, xen) in, closed-platform drivers (vmware, hyperv) out? Drivers for platforms with large (possibly non-OpenStack) production deployments (libvirt, xen, vmware, hyperv) in, drivers without (e.g. docker), out? I think this is in response to demand, not necessarily desire by the nova-core folks. IMHO, maintaining any sort of stable virt driver API for out-of-tree drivers is something we should try to avoid if at all possible. I think the potential option for having a driver moved out of tree would be because the maintainers of that driver would prefer the freedom of merging anything they want without waiting for reviews. As was mentioned earlier in the thread, however, there is a goal to get every driver that is in-tree to have functional testing by Icehouse. This is unrelated to a move for maintenance reasons, and is something I fully support. If we don't have functional testing on a driver, we should consider it broken (and not supported) IMHO. --Dan ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
Re: [openstack-dev] stalled bug fixes for vmware driver
On Sat, Sep 21, 2013 at 6:24 AM, Dan Wendlandt d...@nicira.com wrote: On Fri, Sep 20, 2013 at 1:09 PM, Joe Gordon joe.gord...@gmail.com wrote: On Fri, Sep 20, 2013 at 11:52 AM, Russell Bryant rbry...@redhat.com wrote: On 09/20/2013 02:02 PM, Dan Wendlandt wrote: I couldn't agree more. In fact, the VMware team has been working hard to get a fully-automated CI infrastructure setup and integrated with upstream Gerrit. We already run the tempest tests internally and have been manually posting tempest results for some patches. I wouldn't want to speak for the dev owner, but I think within a very short time (before Havana) you will begin seeing automated reports for tempest tests on top of vSphere showing up on Gerrit. I agree that this will really help core reviewers gain confidence that not only does the code look OK, but that it works well too. How are you doing this? Joshua Hesketh has been working on integrating our internal DB CI tests into upstream zuul, so I wonder there are synergies that can be harnessed here. Michael -- Rackspace Australia ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
Re: [openstack-dev] stalled bug fixes for vmware driver
On Sep 20, 2013 1:27 PM, Dan Wendlandt d...@nicira.com wrote: On Fri, Sep 20, 2013 at 1:09 PM, Joe Gordon joe.gord...@gmail.com wrote: On Fri, Sep 20, 2013 at 11:52 AM, Russell Bryant rbry...@redhat.com wrote: On 09/20/2013 02:02 PM, Dan Wendlandt wrote: I think the real problem here is that in Nova there are bug fixes that are tiny and very important to a particular subset of the user population and yet have been around for well over a month without getting a single core review. Take for example https://review.openstack.org/#/c/40298/ , which fixes an important snapshot bug for the vmwareapi driver. This was posted well over a month ago on August 5th. It is a solid patch, is 54 new/changed lines including unit test enhancements. The commit message clearly shows which tempest tests it fixes. It has been reviewed by many vmware reviewers with +1s for a long time, but the patch just keeps having to be rebased as it sits waiting for core reviewer attention. Personally I tend not to review many vmwareapi patches because without seeing any public functional tests or being able to run the patch myself, I am uncomfortable saying it 'looks good to me'. All I can do is make sure the code looks pythonic and make no assessment on if the patch works or not. With no shortage of patches to review I tend to review other patches instead. I while back Russell announced we would like all virt drivers have a public functional testing system by the release of Icehouse ( http://lists.openstack.org/pipermail/openstack-dev/2013-July/011280.html). Public functional testing would allow me to review vmwareapi patches with almost the same level of confidence as with a driver that we gate on and that I can trivially try out, such as libvirt. Until then, if we put an explicit comment in the release notes explicitly saying that vmwareapi is a group C virt driver (https://wiki.openstack.org/wiki/HypervisorSupportMatrix -- These drivers have minimal testing and may or may not work at any given time. Use them at your own risk) that would address my concerns and I would be happy to +2 vmwareapi patches based just on if the code looks correct and not on how well the patch works. Hi Joe, I couldn't agree more. In fact, the VMware team has been working hard to get a fully-automated CI infrastructure setup and integrated with upstream Gerrit. We already run the tempest tests internally and have been manually posting tempest results for some patches. I wouldn't want to speak for the dev owner, but I think within a very short time (before Havana) you will begin seeing automated reports for tempest tests on top of vSphere showing up on Gerrit. I agree that this will really help core reviewers gain confidence that not only does the code look OK, but that it works well too. Dan Awesome, when that happens I hope to review more vmwareapi patches. Part of the trick will be in how I can see that after a nova patch is merged the vmwareapi system will cover that case going forward. To me, the high-level take away is that it is hard to get new contributors excited about working on Nova when their well-written and well-targeted bug fixes just sit there, getting no feedback and not moving closer to merging. The bug above was the developer's first patch to OpenStack and while he hasn't complained a bit, I think the experience is far from the community behavior that we need to encourages new, high-quality contributors from diverse sources. For Nova to succeed in its goals of being a platform agnostic cloud layer, I think this is something we need a community strategy to address and I'd love to see it as part of the discussion put forward by those people nominating themselves as PTL. I've discussed this topic quite a bit in the past. In short, my approach has been: 1) develop metrics 2) set goals 3) track progress against those goals The numbers I've been using are here: http://russellbryant.net/openstack-stats/nova-openreviews.html Right now we're running a bit behind the set goal of keeping the average under 5 days for the latest revision (1st set of numbers), and 7 days for the oldest revision since the last -1 (3rd set of numbers). However, Nova is now (and has been since I've been tracking this) below the average for all OpenStack projects: http://russellbryant.net/openstack-stats/all-openreviews.html Review prioritization is not something that I or anyone else can strictly control, but we can provide tools and guidelines to help. You can find some notes on that here: https://wiki.openstack.org/wiki/Nova/CoreTeam#Review_Prioritization There is also an aspect of karma involved in all of this that I think phttp://summit.openstack.org/cfp/details/4lays a big part when it comes to the vmware driver patches. To get review attention, someone has to *want* to review it. To get people to want to review your stuff, it can
Re: [openstack-dev] stalled bug fixes for vmware driver
On 09/20/2013 04:11 PM, Dan Wendlandt wrote: Its great that you have dashboards like this, very cool. The interesting thing here is that the patches I am talking about are not waiting on reviews in general, but rather core review. They have plenty of reviews from non-core folks who provide feedback (and they keep getting +1'd again as they are rebased every few days). Perhaps a good additional metric to track would be be items that have spent a lot of time without a negative review, but have not gotten any core reviews. I think that is the root of the issue in the case of the reviews I'm talking about. I feel the pain. Especially when you have a +2 but need to rebase before the second +2 comes in, and lose it. I just sent a pull request to next-review to add --onlyplusone and --onlyplustwo to next-review, to give core reviewers an easy way to focus on already-somewhat-vetted reviews, and leave the new reviews to non-core reviewers. -- David Ripton Red Hat drip...@redhat.com ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
Re: [openstack-dev] stalled bug fixes for vmware driver
On Fri, Sep 20, 2013 at 1:58 PM, Joe Gordon joe.gord...@gmail.com wrote: On Sep 20, 2013 1:27 PM, Dan Wendlandt d...@nicira.com wrote: On Fri, Sep 20, 2013 at 1:09 PM, Joe Gordon joe.gord...@gmail.com wrote: On Fri, Sep 20, 2013 at 11:52 AM, Russell Bryant rbry...@redhat.com wrote: On 09/20/2013 02:02 PM, Dan Wendlandt wrote: I think the real problem here is that in Nova there are bug fixes that are tiny and very important to a particular subset of the user population and yet have been around for well over a month without getting a single core review. Take for example https://review.openstack.org/#/c/40298/ , which fixes an important snapshot bug for the vmwareapi driver. This was posted well over a month ago on August 5th. It is a solid patch, is 54 new/changed lines including unit test enhancements. The commit message clearly shows which tempest tests it fixes. It has been reviewed by many vmware reviewers with +1s for a long time, but the patch just keeps having to be rebased as it sits waiting for core reviewer attention. Personally I tend not to review many vmwareapi patches because without seeing any public functional tests or being able to run the patch myself, I am uncomfortable saying it 'looks good to me'. All I can do is make sure the code looks pythonic and make no assessment on if the patch works or not. With no shortage of patches to review I tend to review other patches instead. I while back Russell announced we would like all virt drivers have a public functional testing system by the release of Icehouse ( http://lists.openstack.org/pipermail/openstack-dev/2013-July/011280.html). Public functional testing would allow me to review vmwareapi patches with almost the same level of confidence as with a driver that we gate on and that I can trivially try out, such as libvirt. Until then, if we put an explicit comment in the release notes explicitly saying that vmwareapi is a group C virt driver ( https://wiki.openstack.org/wiki/HypervisorSupportMatrix -- These drivers have minimal testing and may or may not work at any given time. Use them at your own risk) that would address my concerns and I would be happy to +2 vmwareapi patches based just on if the code looks correct and not on how well the patch works. Hi Joe, I couldn't agree more. In fact, the VMware team has been working hard to get a fully-automated CI infrastructure setup and integrated with upstream Gerrit. We already run the tempest tests internally and have been manually posting tempest results for some patches. I wouldn't want to speak for the dev owner, but I think within a very short time (before Havana) you will begin seeing automated reports for tempest tests on top of vSphere showing up on Gerrit. I agree that this will really help core reviewers gain confidence that not only does the code look OK, but that it works well too. Dan Awesome, when that happens I hope to review more vmwareapi patches. Part of the trick will be in how I can see that after a nova patch is merged the vmwareapi system will cover that case going forward. Great. By the way, all of this automated tempest testing is running nested on top of a physical OpenStack on vSphere cloud we run internally. That cloud has the ability to host labs that are externally accessible. So if you're a core Nova reviewer (or other person who does a lot of Nova reviews), I could definitely look into how we can get you access to a devstack + vSphere environment for your personal use of reviewing + testing patches. Anything I can do to make a core reviewers life easier here, I'm all for. Feel free to reach out to me off-list. Dan To me, the high-level take away is that it is hard to get new contributors excited about working on Nova when their well-written and well-targeted bug fixes just sit there, getting no feedback and not moving closer to merging. The bug above was the developer's first patch to OpenStack and while he hasn't complained a bit, I think the experience is far from the community behavior that we need to encourages new, high-quality contributors from diverse sources. For Nova to succeed in its goals of being a platform agnostic cloud layer, I think this is something we need a community strategy to address and I'd love to see it as part of the discussion put forward by those people nominating themselves as PTL. I've discussed this topic quite a bit in the past. In short, my approach has been: 1) develop metrics 2) set goals 3) track progress against those goals The numbers I've been using are here: http://russellbryant.net/openstack-stats/nova-openreviews.html Right now we're running a bit behind the set goal of keeping the average under 5 days for the latest revision (1st set of numbers), and 7 days for the oldest revision since
Re: [openstack-dev] stalled bug fixes for vmware driver
btw, thanks to the core devs who went and took a look at several of the vmware reviews today. It was like christmas for the team today :) On Fri, Sep 20, 2013 at 2:25 PM, Michael Still mi...@stillhq.com wrote: On Sat, Sep 21, 2013 at 7:12 AM, Dan Wendlandt d...@nicira.com wrote: On Fri, Sep 20, 2013 at 2:05 PM, Michael Still mi...@stillhq.com wrote: How are you doing this? Joshua Hesketh has been working on integrating our internal DB CI tests into upstream zuul, so I wonder there are synergies that can be harnessed here. We're just using the standard stuff built by the OpenStack CI team for third-party testing: http://ci.openstack.org/third_party.html Is that what you were asking, or am I misunderstanding? Ahhh, so that's how our initial prototype was built as well, but the new zuul way is much nicer (he says in a handwavey way). I didn't do the work though, so I can't be too specific apart from saying you don't need to do any of the talking-to-gerrit bits any more -- its possible to just run up a zuul instance which hooks into the upstream one and that runs your tests. zuul handles detecting new reviews and writing results to gerrit for you. Michael -- Rackspace Australia ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev -- ~~~ Dan Wendlandt Nicira, Inc: www.nicira.com twitter: danwendlandt ~~~ ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
Re: [openstack-dev] stalled bug fixes for vmware driver
On 09/20/2013 07:00 PM, Dan Wendlandt wrote: btw, thanks to the core devs who went and took a look at several of the vmware reviews today. It was like christmas for the team today :) And for the record, I did all of my reviews before this thread even started, just as a part of my normal workflow. I was working from the havana-rc1 bug list. :-) -- Russell Bryant ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev